Branch networks play an important role in propagating the effects of bank failure. Firms with fewer branches in their locality are significantly more affected by bank failure, in line with the relationship lending literature that emphasizes the role of "soft" information and physical proximity.

We also show that bank failure is significantly more harmful for the smallest firms in our dataset, in line with the idea that smaller firms face significantly higher switching costs when changing banks than larger firms. We then turn our attention to branch networks.

We first show that the negative effects of bank failures happen both during financial crises AND in normal times. This gives a lot of external validity to important prior literature which presents evidence from crisis periods (e.g.
Chodorow-Reich 2014 & Huber 2018).

To answer this question and to interrogate mechanisms, we first run our estimation for each bank failure separately and find our second surprising result: not all failures are created equal!

For survivors, the pain continues. Firms whose banks fail see their employment growth fall by about 25% compared to similar firms whose banks do not fail. Over a ten-year horizon, these businesses show 6 pp (29%) lower growth rates. No evidence of recovery or catching up.

The long time coverage of our data allows us to study both the short- and long-term (up to a decade!) effects of bank failure. We find that firms affected by bank failure are 6.7 pp (44%) more likely to fail within five years.

After years of data collection and processing, we've assembled what we believe to be the most comprehensive dataset on the banking relationships of small businesses in the US: 36 million loan records spanning three decades, parsed and linked using novel AI methods.

New here and happy to share my job market paper!
"Bad Bank, Bad Luck? Evidence from 1 Million Firm-Bank Relationships"

We build a large novel dataset on US firm-bank relationships to ask the question “How do bank failures affect small business survival and employment?”

A🧵summarizing our findings: